netNMF-sc

netNMF-sc applies network-regularized non-negative matrix factorization to single-cell RNA sequencing (scRNA-seq) data to impute dropout-affected counts and produce biologically informed low-dimensional representations for downstream analysis.

Key Features:

• Network-Regularized Non-Negative Matrix Factorization: Integrates prior gene-gene interaction networks into NMF to enforce proximity of interacting genes in the low-dimensional representation.
• Imputation and Dimensionality Reduction: Imputes gene abundance for zero and nonzero counts while reducing dimensionality and preserving biological signal.
• Clustering Capability: Uses the learned low-dimensional representation to cluster cells into subpopulations for identification of cell types or states.
• Gene-Gene Covariance Estimation: Provides estimates of gene-gene covariance to interrogate regulatory relationships.
• Robustness to Input Network Variations: Delivers reliable results across variations in the input gene interaction network while achieving greater gains with more accurate networks.

Scientific Applications:

• Enhanced Clustering Performance: Improves clustering accuracy compared with existing methods, particularly at high dropout rates (e.g., >60%).
• Estimation of Regulatory Structure: Enables inference of gene-gene covariance and regulatory relationships from scRNA-seq data.
• Dropout Correction for Downstream Analysis: Imputes dropout-affected counts to support downstream analyses such as differential expression and cell-type identification.

Methodology:

Learns a low-dimensional representation of scRNA-seq transcript counts by applying non-negative matrix factorization with network regularization using a prior gene-gene interaction network to keep interacting genes proximal in the reduced space.